I had fun exploring my initial project idea to identify two people on camera using ml5’s BodyPose and BodySegmentation models, and create generative visual and audio effects when the two bodies interacted. This idea was abandoned when I realized how often I was working on developing the code without another body present to help me test it!
I combined the BodyPose and BodySegmentation models into a sketch, added some crude debugging tools / console.logs to help me figure out my next steps. I hope to pick this idea back up when I can develop it with other bodies around to test more frequently!
I’ve recently pivoted on my plans for my thesis project, and after speaking to Dan about it and how I could use the material we’ve been covering in recent weeks to explore some ideas further, he suggested to check out his examples of Calculating Image Similarity, using MobileNet and TensorFlow.js
I started with running the model locally to see what I could learn about the dataset via the results. Dan cautioned that because it’s a small sample of the MobileNet images, the results might be a bit wacky. The results were definitely a bit mixed, but were overall better than I expected.
